June 28, 2026 · 5 min read
Is dbt Worth It for a Marketing Team's Data Stack, Honestly
I've stood up dbt for marketing data three times now, and twice it was the right call. The one time it wasn't is the more useful story.
Every marketing analytics consultant I know has a dbt opinion, and most of those opinions were formed on one project and never updated. Mine comes from three, so let me actually answer the question instead of repeating the hype.
Where it earned its keep
A client had five marketing data sources — GA4 export, ad platform APIs, CRM, email platform, a spreadsheet someone maintained for offline event attribution — all landing in BigQuery as raw tables nobody trusted. The transformation logic that turned those into a usable 'marketing_performance' table lived in eleven different Looker Studio calculated fields, each slightly different, each wrong in a different way. Moving that logic into dbt models with tests wasn't a nice-to-have, it was the only way to get one number everyone agreed on.
dbt doesn't fix bad data. It fixes the fact that five people were each quietly fixing it differently.
Where it was overkill
A different client had two data sources and a part-time marketing ops person. I recommended dbt anyway, because I'd just had a good experience with it, and that was a mistake. The transformation logic was maybe forty lines of SQL. Standing up a dbt project, a Git repo, CI checks, and a deployment job for forty lines of SQL that one person touches twice a quarter added process cost that nobody there had the bandwidth to maintain. Six months later the dbt project was untouched and the useful logic had migrated back into a scheduled query, which is honestly fine.
- Three or more data sources feeding one reporting layer: dbt earns its setup cost almost immediately.
- A team that will actually own the repo, not just the output: without that, dbt becomes one more undocumented thing.
- Transformation logic duplicated across tools (Looker Studio fields, scheduled queries, ad-hoc notebooks): dbt's real value is collapsing that duplication into one tested place.
- A single analyst and simple joins: a scheduled query or two is faster to build and cheaper to maintain.
The honest answer is dbt is worth it exactly when the mess it's solving is bigger than the process it introduces. Measure the mess first. I didn't, the second time, and paid for it in a project nobody used.